Outcome-Based Estimation

The billable hour is a construct inherited from professional services built on manual labour. A solicitor's hour, a consultant's hour, a contractor's hour — each represents a unit of attention applied to a problem. The logic holds as long as effort and output maintain a linear relationship: more hours in, more progress out.

Coding agents have broken that linearity.

A senior engineer directing Cursor, GitHub Copilot, or Claude Code can produce a complete, tested authentication module in forty minutes that would have taken a junior developer four days. The input — time — has collapsed. The output — functional software — has not. The ratio between effort and outcome has fundamentally shifted. Yet most engineering engagements continue to price the effort rather than the outcome.

This is the Time-Tax Trap: charging for the mechanism when the client is purchasing the result.

The structural problem this creates is Billability Theatre — a condition in which engineering progress is measured by logged hours rather than verified functional outcomes. A team can spend forty hours producing a landing page that a developer equipped with modern agent tooling can produce in four. The client receives an identical asset. The time delta is pure extraction from the engagement. It produces no additional value; it simply inflates the cost of the mechanism.

Traditional hour-based models were designed to manage risk for the consultant, not to deliver capital efficiency for the founder. In a manual-labour environment, the hour was the only observable proxy for work in progress. In an AI-native environment, it is simply the wrong unit of measurement.

The cost is not any individual invoice. The cost is the structural incentive the hour creates: the longer it takes, the more the consultant earns. Agent-Assisted Velocity inverts this. The faster TechTek delivers a verified outcome, the better the outcome for the client — and the better the signal about the team's architectural efficiency. A fee model built on hours actively penalises this efficiency. The OBE model rewards it.

The Outcome-Based Estimation model replaces the hour with the Verified Outcome. A project is decomposed into discrete, auditable functional assets — each defined not by the work required to produce it, but by the state transition it represents in the product.

A state transition reads: "User authentication does not exist → Functional auth layer, deployed, tested, accepted." The fee attaches to the transition. Not to the minutes spent traversing it.

This is Labour-Logic Decoupling: structurally disconnecting the fee from the manual labour input and attaching it to the verified functional output. It is not a pricing preference — it is a governance decision that changes the incentive structure of the entire engagement.

Three mechanisms make the OBE model operational:

The pipeline has one structural property the legacy model lacks: every step produces evidence. The Classifier produces a written complexity score. The Commercial Lock commits both parties to measurable success criteria before a line of code is written. Each Synthesis Loop sprint produces a deployable artefact. The Measurement gate produces a binary pass/fail on objective data — triggering the variable fee on outcomes, not assertions. At no point does the client have to take the team's word for progress.

The OBE Classifier routes each piece of work into one of two channels based on the architectural guidance required to produce a Verified Outcome. The channel determines who leads, which tools are used, and what the timeline looks like — not the complexity of the code itself.

The distinction is not about the volume of code — it is about the irreducibility of the judgment required. Commodity logic can be guided to completion by a well-prompted agent. Core business logic requires a human who can reason about the problem space, evaluate trade-offs the agent cannot see, and make architectural decisions that will outlast the current sprint.

The most common misconception about agentic engineering is that higher generation speed reduces the need for senior oversight. The opposite is true. A coding agent can produce five hundred lines of structurally plausible code in ninety seconds — code that compiles, passes linting, and looks correct on inspection. That same code may contain architectural decisions that accumulate Hidden Technical Debt at a rate no manual team could match, precisely because the rate of generation is so high.

The governance positions below are not optional constraints on the OBE model. They are load-bearing elements of it. Remove any one of them and the model produces fast-generated software that is technically unsound and commercially misaligned.